What it is
A model tuned on your domain and your documents, running where your policies require - your cloud tenant or your own hardware. For repeat work in a narrow domain, you stop renting general intelligence by the token and start running a model that already knows the territory.
When you need it
On-prem or policy limits on cloud LLMs, a domain that needs hardened outcomes, or repeat work where owning the model cuts token spend.
What a first build looks like
A tuned model hardened on one report domain where reliability matters most - including Microsoft's frontier tuning path on Azure.